krites doctor¶
What krites doctor does¶
Run diagnostic checks over the Go runtime, the resolved configuration, credential storage and file permissions, then print a per-check pass / warn / fail / skip report. Run it when krites misbehaves, to pinpoint a misconfigured or missing dependency before reading logs.
Usage¶
Flags¶
| Flag | Default | What it does |
|---|---|---|
--fail-on |
warn under CI, fail otherwise |
Exit non-zero when any check is this bad or worse: none, fail or warn |
Plus the global flags on every command.
Reading the output¶
Each line is one check with its verdict:
krites v0.10.1
[OK] Go version: go1.27.1
[OK] Configuration: loaded from /home/you/.krites/config.yaml
[OK] Credential storage: no literal credentials in config
[OK] Permissions: config dir: /home/you/.krites (-rwx------)
[OK] Forge adapters: GitLab linked
[OK] Release source: gitlab provider linked
[OK] Chat providers: claude, gemini linked for the tool's own use (ai feature off)
A skip is not a failure. It means the check does not apply to how krites is
set up: a provider that is off has nothing to test, and on a fresh install with
no config file yet, doctor reports that rather than creating one.
Forge adapters, Release source and Chat providers report what the
binary links: GitLab for self-update, and Claude and Gemini for
krites review. A missing one means a build that can't update or
review.
krites doctor report: a support bundle ready to paste¶
Where plain doctor gives a health verdict, doctor report adds the state dump
around it: tool and runtime versions, the resolved configuration, the config
paths that contributed, feature-flag state, and the full doctor report.
The whole bundle is secret-redacted before it is written, in both text and
JSON form, and there is no option to disable the redaction. If you need a raw
value, read that specific key with krites config get <key> --unmask.
krites doctor report --output json gives the same bundle as JSON, for attaching
to an issue.
What doctor does not check¶
doctor validates the environment and configuration. It does not:
- reach a configured AI provider or prove its API key works;
- load the ONNX Runtime or the face/removal models;
- open or validate a shoot.
For a provider that is configured but failing, run the operation itself. The
studio's Review action reports its own availability, and
krites cull fails loudly when a configured face provider cannot
start.